Insights on enterprise AI, evaluation, governance and financial inclusion

425 pages, nine chapters, four hundred citations — distilled into a reference you can scan in fifteen minutes.

Why it matters: The Index is the most-cited, least-read document in enterprise AI. This is the map I send clients before the next steerco.

When enterprise AI fails, it usually fails above the model — in trust, accountability and the messy business of actually running the thing.

Why it matters: Assume the data is secure, the model accurate and budget unlimited, and most agentic programmes still stall. The remaining failure modes are structural, and they are the ones nobody has bought a tool for.

The models that hurt you are, almost by definition, the ones your inventory does not count.

Why it matters: The new wave of AI-risk regulation is aimed at a much older problem than AI: institutions do not know which decision systems they are running.

Every AI dashboard was green. Delivery was slower.

Why it matters: The productivity metric that flatters your AI tooling — and the one that would tell you the truth. A field story, not a survey.

We stopped accumulating things and started accumulating access. AI may quietly reverse that.

Why it matters: A slower-moving essay on how AI reshapes the consumer preference between owning and subscribing — with implications for durable-goods, media and platform strategy.

Both the bears and the bulls are answering a question the industry stopped facing in 2022.

Why it matters: The real disruption to Indian IT services is older, slower, and already inside the home — not the LLM headline you think.

The Indian OTA stack is a structural failure. Generative and agentic AI can meet it — if anyone builds for the buyer, not the funnel.

Why it matters: A worked example of where an AI-native product opportunity actually sits in a stagnant category.

Same goal. Very different journeys — and the need for control has never been greater.

Why it matters: The delivery methodology arc, and why AI-native does not mean less governance. It means governance in different places.

41% of code generated. A fraction survives to production.

Why it matters: The gap between developer consumption and business value is where the AI coding ROI conversation actually lives.

The AI J-curve isn't a tech problem — it's an organisational-design problem.

Why it matters: Why the productivity dip is structural, and what to redesign (roles, incentives, review loops) before expecting the curve to turn up.

A prompt instruction is not a security control. It is a wish, written on a sticky note and taped to a vault.

Why it matters: Most enterprise agents that already read customer records and move money are guarded by nothing sturdier than a politely worded instruction.

One vendor's score becomes everyone's verdict.

Why it matters: The largest audit of hiring AI to date shows monoculture — not bias — is the real systemic risk. Read before your next vendor consolidation memo.

Scale is the easy problem. Construct validity is the one no smarter model will fix.

Why it matters: Two failure modes of large-scale assessment survive every model upgrade. Frame this before betting a curriculum on autograders.

The evaluation didn't fail. It finally told the truth.

Why it matters: A read on what a public evaluation collapse reveals about the gap between what we test for and what we actually value.

The AI delusion that nearly destroyed one man is a preview of how it happens to organisations — at scale, with a budget, and no one left to pull the plug.

Why it matters: Individual model-induced psychosis has an organisational analogue. Governance implication: who owns the off-switch when the pilot is going well?

A switch you never test is just a battery quietly going flat.

Why it matters: Regulators now mandate kill switches for AI systems. Turning a system off is the easy part; surviving the state you are in once it is off is the whole game.

Sovereignty used to be about territory. Increasingly it is about inference.

Why it matters: Why data-residency debates are the small version of a much bigger question about who controls the models a country depends on.

The pixels got cheap. The editability didn't.

Why it matters: A field note on where the AI production line is quietly billing everyone downstream — and why 'ship fast' has an externality nobody is booking.

The poor don't pay more for money. They pay more for being verified.

Why it matters: The companion essay to my Responsible AI for Financial Inclusion talk at AINext Dubai. Reframes inclusion as a verification-cost problem AI can actually address.